Safety Loading and Exponential Ruin Decay

Why Safety Loading Affects the Ruin Bound

In a classical surplus model, premiums must cover expected claims and provide a margin for adverse fluctuations. That margin is the safety loading. Its effect is not limited to increasing the average surplus: it also changes the adjustment coefficient, which determines how quickly the standard upper bound on ultimate ruin decreases as initial capital grows.

The adjustment coefficient is not itself a ruin probability. It is a positive rate in the exponent of an upper bound. For a fixed claim severity model, a higher safety loading gives a larger adjustment coefficient whenever the corresponding positive roots exist, and therefore a smaller Lundberg bound at any positive initial capital.

Definition: In the classical compound Poisson surplus model, let $u\geq 0$ be initial capital, let $N(t)$ be a Poisson process with rate $\lambda$, and let $X_1,X_2,\ldots$ be independent, identically distributed nonnegative claim amounts, independent of $N(t)$. Write $S(t)=\sum_{j=1}^{N(t)}X_j$ and $U(t)=u+ct-S(t)$, where $c$ is the premium rate. If $E[X]$ is finite and positive, the premium rate is written $c=(1+\theta)\lambda E[X]$, where $\theta>0$ is the safety loading. Ultimate ruin has probability $\psi(u)=P(\text{\(U(t)<0$ for some }t\geq 0)\).

Let $M_X(r)=E[e^{rX}]$ be the moment-generating function of claim severity, finite at the values of $r$ under consideration. The adjustment coefficient $R$ is a positive solution of the Lundberg equation. In terms of the safety loading, the equation is

\lambda\bigl(M_X(R)-1\bigr)=cR \qquad\Longleftrightarrow\qquad \frac{M_X(R)-1}{R}=(1+\theta)E[X].

The second form follows by substituting $c=(1+\theta)\lambda E[X]$ and dividing by $\lambda R$, which is positive. It makes the role of loading especially visible: for fixed severity, increasing $\theta$ raises the right-hand side.

Theorem: If the classical model satisfies the net profit condition $c>\lambda E[X]$ and has a positive adjustment coefficient $R$, Lundberg’s inequality gives $\psi(u)\leq e^{-Ru}$ for $u\geq 0$. The bound uses $R$ in inverse units of claim amount, so $Ru$ is dimensionless. It is an upper bound, not generally an exact value of $\psi(u)$.

How Loading Changes the Adjustment Coefficient

Keep $\lambda$ and the claim severity distribution fixed, so that only the premium rate changes with $\theta$. The positive root of the Lundberg equation moves to the right as the loading increases, provided a positive root exists for each loading being compared. This monotonicity explains why a larger loading tightens the exponential upper bound at a fixed $u>0$.

Write the root equation.
For each loading, $R$ satisfies $(M_X(R)-1)/R=(1+\theta)E[X]$.

Compare the right-hand sides.
If $\theta_2>\theta_1$, then $(1+\theta_2)E[X]>(1+\theta_1)E[X]$, since $E[X]>0$.

Use the increasing left-hand side.
For nonnegative, nondegenerate $X$, $(M_X(r)-1)/r$ is strictly increasing for $r>0$ wherever the moment-generating function is finite. Thus the larger right-hand side requires a larger positive root.

Translate the root change into a bound.
If $R_2>R_1$, then $e^{-R_2u}<e^{-R_1u}$ for $u>0$.

For completeness, the increase in the left-hand side can be seen from

\frac{M_X(r)-1}{r} =E\left[\frac{e^{rX}-1}{r}\right] =E\left[\int_0^X e^{rt}\,dt\right].

For each $X>0$, the integrand $e^{rt}$ increases with $r$, so the integral increases with $r$; taking expectations preserves that ordering. This argument also shows why the claim distribution must be held fixed for this comparison. If the severity distribution changes, the left-hand side changes too.

A positive root is not guaranteed for every possible severity distribution and loading. The moment-generating function must be finite where the equation is evaluated, and the equation must actually have a positive solution. Therefore, the comparison is conditional: when both roots exist for the fixed severity model, the larger loading has the larger root. The net profit condition alone does not guarantee that an adjustment coefficient exists.

Worked Calculations

Worked Example: Exponential Claims and Two Loadings

A hypothetical insurer has Poisson claim frequency $\lambda=2$ claims per year and exponential claim amounts with mean $\beta=\$10{,}000$. Compare the adjustment coefficient and Lundberg bound at initial capital $u=\$50{,}000$ for safety loadings $\theta=0.25$ and $\theta=0.50$. Assume the classical surplus model and round probabilities to four decimals.

Method. For an exponential claim amount with mean $\beta$, $M_X(r)=1/(1-\beta r)$ for $0\leq r<1/\beta$. Its mean is $\beta$, and the adjustment equation becomes

\lambda\left(\frac{1}{1-\beta R}-1\right) =(1+\theta)\lambda\beta R.

For the positive root, cancel $\lambda R$ and solve:

\frac{\beta}{1-\beta R}=(1+\theta)\beta, \qquad R=\frac{\theta}{(1+\theta)\beta}.

Loading $\theta=0.25$. The premium rate is $c=1.25(2)(\$10{,}000)=\$25{,}000$ per year. The adjustment coefficient is

R=\frac{0.25}{1.25(10{,}000)} =0.000020\text{ per dollar} =2.0000\times 10^{-5}\text{ per dollar}.

The exponent is $Ru=(0.000020)(50{,}000)=1$, so the bound is $e^{-1}=0.3679$.

Loading $\theta=0.50$. Now $c=1.50(2)(\$10{,}000)=\$30{,}000$ per year, and

R=\frac{0.50}{1.50(10{,}000)} =\frac{1}{30{,}000} =3.3333\times 10^{-5}\text{ per dollar}.

Here $Ru=50{,}000/30{,}000=5/3$, giving the bound $e^{-5/3}=0.1889$. The higher loading has increased $R$, and the bound at the same capital is smaller. These calculations give upper bounds, not exact ruin probabilities. Common mistake avoided: The claim frequency cancels from the adjustment equation for exponential severity, but it still enters the premium rate; omitting $\lambda$ when calculating $c$ would give incorrect premium units and amount.

Worked Example: Erlang-2 Claims and an Explicit Root

Suppose a hypothetical portfolio has Poisson frequency $\lambda=4$ claims per year and Erlang-2 claim amounts with scale parameter $\beta=\$5{,}000$, so $E[X]=2\beta=\$10{,}000$. Take $\theta=0.50$ and $u=\$20{,}000$. Find the adjustment coefficient and the Lundberg bound, rounding the probability to four decimals.

Method. For this Erlang-2 model, $M_X(r)=(1-\beta r)^{-2}$ for $0\leq r<1/\beta$. The premium rate is $c=1.50(4)(\$10{,}000)=\$60{,}000$ per year. Set $z=\beta R$, so $0<z<1$. Substitution into the adjustment equation gives

4\left(\frac{1}{(1-z)^2}-1\right) =60{,}000\frac{z}{5{,}000} =12z.

Dividing by 4 yields $(1-z)^{-2}-1=3z$. Multiply by $(1-z)^2$, expand, and use $z>0$:

2z-z^2=3z(1-2z+z^2), \qquad 2-z=3-6z+3z^2, \qquad 3z^2-5z+1=0.

The roots are $(5\pm\sqrt{13})/6$. Since $\sqrt{13}=3.605551$, the smaller root is $z=(5-\sqrt{13})/6=0.232408$. The other root exceeds 1 and is outside the moment-generating function’s domain, so it cannot be used. Thus

R=\frac{z}{\beta} =\frac{0.232408}{5{,}000} =4.6482\times 10^{-5}\text{ per dollar}.

At $u=\$20{,}000$, $Ru=(4.6482\times10^{-5})(20{,}000)=0.9296$, so the ultimate ruin bound is $e^{-0.9296}=0.3947$. Common mistake avoided: The root must lie inside the MGF domain $z<1$; selecting the other algebraic root would produce an invalid adjustment coefficient.

Worked Example: Capital for a Target Bound at Two Loadings

A hypothetical insurer has exponential claim amounts with mean $\$8{,}000$. Compare the minimum initial capital needed to make the Lundberg bound no greater than 0.10 when $\theta=0.50$ and when $\theta=1.00$. The claim frequency is $\lambda=3$ per year. Report capital to the nearest cent.

Method. For exponential severity, $R=\theta/((1+\theta)\beta)$, where $\beta=\$8{,}000$. To make the bound at most 0.10, solve $e^{-Ru}\leq0.10$. Taking logarithms gives $u\geq \ln(10)/R$.

For $\theta=0.50$, the premium rate is $c=1.50(3)(\$8{,}000)=\$36{,}000$ per year and

R=\frac{0.50}{1.50(8{,}000)} =\frac{1}{24{,}000}\text{ per dollar}, \qquad u\geq24{,}000\ln(10)=\$55{,}262.04.

For $\theta=1.00$, the premium rate is $c=2(3)(\$8{,}000)=\$48{,}000$ per year and

R=\frac{1.00}{2(8{,}000)} =\frac{1}{16{,}000}\text{ per dollar}, \qquad u\geq16{,}000\ln(10)=\$36{,}841.36.

The higher loading lowers the capital required by this bound. At either unrounded threshold, $Ru=\ln(10)$, and $e^{-Ru}=0.10$. Common mistake avoided: Solving $e^{-Ru}=0.10$ gives $u=\ln(10)/R$, not $0.10/R$; the logarithm is essential.

Interpreting the Result Carefully

For a fixed severity model, $R$ measures how strongly the ultimate ruin upper bound falls as initial capital increases: multiplying capital by an amount $\Delta u$ multiplies the bound by $e^{-R\Delta u}$. The larger $R$ is, the faster this particular bound decays with capital. It does not follow that the true ruin probability equals the bound or that changing the loading changes severity.

The loading comparison also depends on using a consistent premium convention. Here $\theta$ is the percentage added to expected aggregate claims per unit time, so $c=(1+\theta)\lambda E[X]$. The factor multiplying expected claims is $1+\theta$, not $\theta$. This distinction matters in both the adjustment equation and premium calculations.

Key takeaway: With claim frequency and severity held fixed, increasing the safety loading increases the premium rate and, when the positive roots exist, increases the adjustment coefficient. Since $\psi(u)\leq e^{-Ru}$, the resulting Lundberg upper bound is smaller for every $u>0$.

Check Your Understanding

For each problem, use the classical compound Poisson model and the stated Lundberg bound. Give probabilities to four decimals.

A hypothetical insurer has exponential claims with mean $\$4{,}000$ and safety loading $\theta=0.20$. What is the adjustment coefficient? (Answer: $R=1/24{,}000=4.1667\times10^{-5}$ per dollar.)
Exponential claims have mean $\$12{,}000$, with $\lambda=2$ claims per year and $\theta=0.50$. Find the Lundberg bound at $u=\$24{,}000$. (Answer: $R=1/36{,}000$ per dollar and bound $e^{-2/3}=0.5134$.)
For Erlang-2 claims with scale $\beta=\$3{,}000$ and safety loading $\theta=1.00$, the scaled root $z=\beta R$ satisfies $(1-z)^{-2}-1=4z$. Find the admissible root and $R$. (Answer: $z=(7-\sqrt{17})/8=0.3596$, so $R=1.1987\times10^{-4}$ per dollar.)
For exponential claims with mean $\$5{,}000$, compare $R$ at loadings 0.25 and 0.50. By what factor does the adjustment coefficient increase? (Answer: $R_{0.25}=4.0000\times10^{-5}$, $R_{0.50}=6.6667\times10^{-5}$ per dollar; the factor is $5/3$.)
For exponential claims with mean $\$10{,}000$ and $\theta=0.25$, find the least capital for which the Lundberg bound is at most 0.20. (Answer: $R=1/50{,}000$ per dollar and $u\geq50{,}000\ln(5)=\$80{,}471.90$.)

What You'll Learn